Research map / Label-efficient and robust learning / Recognition and long-tail
Self-Supervised Aggregation of Diverse Experts for Test-Agnostic Long-Tailed Recognition
Yifan Zhang, Bryan Hooi, Lanqing HONG, Jiashi Feng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use skill-diverse experts trained with cross-entropy, balanced softmax, and inverse softmax | SADE · ImageNet-LT, uniform test | top-1 accuracy (All): 56.3 -> 58.8 | Improved |
| Add test-time self-supervised expert aggregation (ablation) | SADE · ImageNet-LT, Forward-LT-50 | top-1 accuracy (All): 65.5 -> 69.4 (+3.9) | Improved |
| Add test-time self-supervised expert aggregation (ablation) | SADE · ImageNet-LT, Uniform | top-1 accuracy (All): 58.8 -> 58.8 (+0.0) | No clear effect |
| Add test-time self-supervised expert aggregation (ablation) | SADE · ImageNet-LT, Backward-LT-50 | top-1 accuracy (All): 49.8 -> 53.1 (+3.3) | Improved |
| Add test-time self-supervised expert aggregation (ablation) | SADE · ImageNet-LT, only few-shot classes | top-1 accuracy: 42.5 -> 53.6; +11.1 | Improved |
| Replace learned expert-specific objectives with independently trained cross-entropy experts as in RIDE (ablation) | SADE expert ensemble · ImageNet-LT, uniform test | top-1 accuracy (All): 58.8 vs 56.3 | Hurt |
| Use prediction-stability aggregation rather than test-time pseudo-labeling | SADE · ImageNet-LT, Backward-LT-50 | top-1 accuracy: 42.4 -> 53.1 | Improved |
| Use prediction-stability aggregation rather than entropy minimization with Tent | SADE · ImageNet-LT, Backward-LT-50 | top-1 accuracy: 45.7 -> 53.1 | Improved |
| Train with stronger RandAugment for 400 epochs (ablation) | SADE · ImageNet-LT, uniform test | accuracy: 58.8 -> 61.2 | Improved |
Closest papers by meaning
- Harnessing Hierarchical Label Distribution Variations in Test Agnostic Long-tail Recognition (ICML 2024 · similarity 0.89)
- A Square Peg in a Square Hole: Meta-Expert for Long-Tailed Semi-Supervised Learning (ICML 2025 · similarity 0.89)
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · similarity 0.88)
- Breaking Long-Tailed Learning Bottlenecks: A Controllable Paradigm with Hypernetwork-Generated Diverse Experts (NeurIPS 2024 · similarity 0.87)
- Three Heads Are Better than One: Complementary Experts for Long-Tailed Semi-supervised Learning (AAAI 2024 · similarity 0.87)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
